GeraldNdawula/

deepface_detection_intial_eval_results

This repo contains the results from our Stage 1 baseline deepfake classifier

-# Initial Evaluation Results Baseline ViT runs with different training settings:

epochs=1, batch_size=8, lr=1e-4
epochs=3, batch_size=8, lr=1e-4
epochs=5, batch_size=16, lr=1e-4
epochs=8, batch_size=16, lr=5e-5
epochs=10, batch_size=32, lr=5e-5

Best run:

epochs=8, batch_size=16, lr=5e-5
Accuracy: 0.64 | F1: 0.57 | ROC-AUC: 0.71

See… See the full description on the dataset page: https://huggingface.co/datasets/GeraldNdawula/deepface_detection_intial_eval_results.